|
|
|
 |
|
|
データ可視化,データマイニングおよび統計のための高速な直感的ツール
DataDesk
|
|
Editorial Backgrounder
Why Data Desk emphasizes Exploratory Data Analysis
Data Desk is designed to help the researcher or business person make the best-informed decisions possible. Today's science, industry, and economy are complex -- even overwhelming. We must understand masses of data from a variety of sources in order to make effective decisions.
The modern science of Statistics is about a century old. In its early days, it concentrated on analyses of data, considering effective ways to describe patterns, trends, and relationships. In the middle half of the 20th century, attention moved to developing a solid mathematical foundation, establishing the properties of various estimators in an attempt to find the "best" methods.
In 1962 John Tukey issued a warning that mathematical statistics was ignoring real-world data analysis. Tukey called for a return to scientific statistics in which the value of the statistical description of the data was paramount. In subsequent work, Tukey defined Exploratory Data Analysis, a philosophy that returned to the original goals of statistics, but used modern methods.
EDA was initially promoted by few statisticians, but in recent years it has grown in acceptance. A large part of this growth is due to the availability of desktop computers and the explosion of data for which traditional statistics is just not suitable. Desktop computers have also made it possible to develop new graphical methods that support the EDA philosophy in strikingly effective fashion.
Data Desk implements traditional statistics techniques that are suitable for data from planned experiments and sample surveys. But Data Desk goes further, offering powerful tools for data exploration -- tools that are useful even for data that may not meet the high standards of traditional statistics. These tools use innovative graphics in ways that are natural even for people who are not trained in statistics. Such data exploration not only reveals patterns in real-world data, but also brings to light what doesn't fit -- often the most important discovery in the data.
Most data arise as a byproduct of other activities. For example, a business person may have data in a spreadsheet intended for tracking sales, data in a database for human resource management, or data that have been published by a government or trade organization. A researcher may collect data to sift a variety of alternatives, may want to look in a new way at data originally collected for a different purpose, or may want to check experiment data for errors or unexpected patterns. Data Desk is designed specifically for these functions.
Traditional inferential statistics starts from a hypothesis, performs an experiment, and then tests the hypothesis. EDA starts instead from the data and asks what patterns, relationships, or trends they might hold. As defined originally by Dr. John Tukey, EDA emphasizes data display, finding simple functional descriptions of patterns in the data, and examining the residuals for evidence of deeper patterns or interesting exceptions. Unlike traditional statistics, EDA acknowledges that data are often heterogeneous, and provides effective tools for identifying and isolating extraordinary values and separate subgroups.
Because EDA relies heavily on data display, makes few assumptions about the structure of the data and emphasizes identifying and describing patterns, it is useful to a wide range of professionals who can recognize important patterns easily, but may not wish to work with complex statistical techniques. EDA is the foundation of a growing trend that empowers people who have data and want to discover the patterns hiding within.
|
| DataDeskのTopページに戻る |
|
|
